Cerebras Stock Post-IPO Low 2026: Lockup Expiry vs Nvidia Pressure

Cerebras stock just had its worst week as a public company. CBRS fell nearly 20% and closed Friday at $166.43, its lowest level since the May IPO and below the $185 offer price. Two things hit in the same week. Another wave of insider lockup shares became sellable. Fresh reports also said Nvidia, not Cerebras, will power a key part of OpenAI’s next-generation stack. For anyone holding or watching AI chip stocks, this tests whether a specialist chipmaker can keep a premium valuation while Nvidia goes after its one clear selling point: very fast inference.

What’s actually new with Cerebras stock

Cerebras priced its IPO on May 14, 2026 at $185 a share, well above the marketed range of $150 to $160. It raised $5.55 billion, the largest US IPO of the year so far. The stock opened at $350, briefly touched $385, and closed its first day at $311.07. It has since given back all of those gains and more. It is now down more than 50% from its peak and trades below what institutions paid at the IPO. That last point matters most, because IPO buyers who are underwater tend to sell into any bounce.

The Cerebras IPO lockup expiration is not a single event. It is a staggered schedule of roughly a dozen tranches. Large blocks unlocked after the Q1 earnings report (about 27.7 million shares) and the Q2 report (about 36.4 million shares), followed by biweekly releases of about 14.6 million shares each. This week’s tranche made about 19.4 million more shares eligible for sale, roughly 8% of the approximately 238 million shares outstanding. Releases of the same size are scheduled for October 14 and October 28. The full lockup ends no later than November 9, or earlier if Cerebras files Q3 earnings first. The IPO float was only about 34.5 million shares, so this is a lot of new supply for a thinly traded stock.

The second hit came from competition. Reports that Nvidia hardware will power a key component of OpenAI’s newest model release weighed on a company whose revenue depends heavily on OpenAI. Nvidia also keeps improving price and performance on low-latency inference, which is exactly where Cerebras’s wafer-scale pitch is strongest.

Why it matters

• The market is now pricing in customer concentration risk. Cerebras gets a significant share of its revenue from OpenAI. Investors treat any sign that OpenAI is sending workloads to Nvidia as a direct threat to future Cerebras inference cloud revenue, even though the reported $25.4 billion backlog (remaining performance obligations as of June 30) has not changed.
• The fundamentals and the stock are telling different stories. Q2 revenue was about $180 million, up 74% year over year. That is real growth, but the peak valuation assumed near-perfect execution. The multiple is now shrinking faster than the business is changing.
• Speed alone is getting harder to defend. The WSE-3 wafer-scale chip keeps model weights in 44GB of on-chip SRAM, which is why Cerebras leads on single-user tokens per second. If Nvidia narrows that gap with Blackwell, Rubin, and its Groq-derived inference technology, Cerebras loses the main reason buyers choose it.
• The supply overhang is mechanical, not a verdict on the business. Insiders and early employees holding low-cost shares will sell to diversify regardless of the outlook. That pressure has a known end date, which is why traders are circling November 9.
• It sets the tone for other AI chip IPOs. AI hardware startups weighing a listing are watching closely. A specialist trading below its offer price makes their roadshows harder.
• Cerebras vs Nvidia inference is now a public-market scorecard. Every benchmark, customer win, and model launch will move CBRS, making it one of the most news-sensitive AI chip stocks.

How to use it today

Investors and builders alike can track this story with free tools and their own tests instead of headlines. This is not investment advice. It is a workflow for making our own decisions.

1. Put the lockup dates on a calendar. The remaining tranches are October 14, October 28, and the final release on or before November 9. Watch volume and price action in the two or three sessions around each date. If the stock holds up on heavy volume, buyers are absorbing the new supply.

   2026-10-14  ~19.4M shares eligible
   2026-10-28  ~19.4M shares eligible
   2026-11-09  remaining locked shares (or earlier on Q3 earnings filing)

2. Pull the price and volume data yourself. A short Python script using the free yfinance library compares daily volume to the 20-day average, so unusual selling around unlock dates stands out. Install the dependencies first:

   pip install yfinance pandas

   Then run:

   import yfinance as yf
   df = yf.download("CBRS", period="6mo", interval="1d")
   df["vol_20d"] = df["Volume"].rolling(20).mean()
   df["vol_ratio"] = df["Volume"] / df["vol_20d"]
   print(df[["Close", "Volume", "vol_ratio"]].tail(15))

3. Watch insider sales through SEC EDGAR. Form 4 filings show when directors and officers actually sell. Being eligible to sell and actually selling are different signals. EDGAR’s free API requires a descriptive User-Agent header.

   curl -s -H "User-Agent: YourName your@email.com" \
     "https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&company=cerebras&type=4&dateb=&owner=include&count=40"

4. Benchmark the product claim yourself. The bull case rests on Cerebras being much faster at inference, and that is easy to test. Cerebras’s inference API is OpenAI-compatible, so we can time the same prompt on Cerebras and on an Nvidia-backed provider. The model name below is a placeholder; check Cerebras’s docs for the current model list.

   curl https://api.cerebras.ai/v1/chat/completions \
     -H "Authorization: Bearer $CEREBRAS_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "model": "MODEL_NAME_FROM_DOCS",
       "messages": [{"role": "user", "content": "Write a 500-word summary of wafer-scale computing."}],
       "max_tokens": 800
     }' -w "\nTotal time: %{time_total}s\n"

   Divide output tokens by generation time to get tokens per second, then run the same test against a GPU-backed endpoint. A shrinking gap matters more than any analyst note.

5. Use an LLM to stress-test the earnings call. When Q3 results land, paste the transcript into any free assistant with a structured prompt:

   You are a skeptical semiconductor analyst. From this earnings transcript, extract:
   1. Revenue and YoY growth
   2. Remaining performance obligations (backlog) vs last quarter
   3. % of revenue from the largest customer
   4. Any comments on Nvidia, pricing pressure, or inference cloud growth
   5. Guidance changes
   Flag anything management avoided answering. Transcript: [PASTE]

How it compares

Cerebras does not compete with all of Nvidia. It competes with Nvidia’s fastest inference products. Here is how the main options compare on the questions buyers care about.

Platform	Architecture	Inference strength	Main weakness	Ecosystem
Cerebras WSE-3 (CS-3)	Single wafer-scale chip, ~4 trillion transistors, 900,000 cores, 44GB on-chip SRAM	Industry-leading single-user tokens per second	Customer concentration; smaller software ecosystem	Own SDK plus an OpenAI-compatible inference cloud
Nvidia Blackwell (B200/GB200/B300)	Multi-die GPUs with HBM, NVLink rack-scale systems	Strong throughput and cost per token at scale	Higher per-user latency than SRAM-heavy designs	CUDA, TensorRT-LLM, Dynamo; the default choice
Nvidia Rubin (Vera Rubin)	Next-gen GPU with HBM4, ramping from late 2026	Big gains in inference performance per watt	Supply allocation; expensive	Same CUDA stack, so switching costs are low for existing buyers
Groq LPU tech (licensed by Nvidia)	SRAM-based deterministic inference chips	Low-latency inference, the same pitch as Cerebras	Now part of Nvidia’s plans rather than an independent rival	Folding into Nvidia’s inference roadmap
AMD Instinct MI355X / MI400	GPU with large HBM capacity	Memory-heavy inference, competitive cost per token	ROCm software still behind CUDA	Growing hyperscaler adoption

Our view: Cerebras still wins on raw speed for interactive and agentic workloads, where every millisecond shapes the user experience. But Nvidia does not need to beat Cerebras on speed. It only needs to get close enough that buyers stay on CUDA.

What’s next for Cerebras stock

The next six weeks are mostly about supply. The October 14 and October 28 tranches and the final November 9 release will keep pressure on the stock until the overhang clears. Stocks with heavy lockup overhangs often bottom around the final release, not before it, though that is a tendency, not a rule. If CBRS holds the mid-$160s through these dates on high volume, sentiment could turn quickly once the forced selling ends.

Q3 earnings will matter more than any headline. The CBRS stock outlook 2026 hinges on three numbers: revenue growth (can it stay near the Q2 pace of 74%?), backlog (does the $25.4 billion RPO grow or shrink?), and customer mix (is non-OpenAI revenue a growing share of the total?). A more diversified customer base would directly answer the bear case. Any sign that OpenAI is cutting its commitment would confirm it.

Further out, Nvidia’s Rubin ramp and its use of the licensed Groq inference technology will shape the competitive picture. Cerebras needs its own roadmap: a WSE-3 successor, more data center capacity for its inference cloud, and broader developer adoption. Watch for new sovereign and enterprise deals, inference pricing changes, and third-party benchmarks. Those signals will show whether Cerebras is a durable specialist or a product feature Nvidia eventually absorbs.

Frequently Asked Questions

Why did Cerebras stock drop 20% this week?

Two things happened at once. About 19.4 million insider and employee shares became eligible for sale under the staggered lockup schedule. Separately, reports that Nvidia will power a key part of OpenAI’s latest model raised concerns about Cerebras’s biggest customer relationship. The stock closed at $166.43, below its $185 IPO price.

When does the Cerebras IPO lockup fully expire?

The final release is scheduled for no later than November 9, 2026, 180 days after the IPO prospectus date. It could come earlier if Cerebras files its Q3 earnings report first. Tranches of about 19.4 million shares are also due on October 14 and October 28.

Is Cerebras faster than Nvidia for inference?

For single-user, low-latency inference, the WSE-3 has generally led published tokens-per-second benchmarks because model weights sit in on-chip SRAM instead of external HBM. Nvidia usually competes on total throughput, cost per token at scale, and its software ecosystem. The gap is narrowing as Nvidia prioritizes fast inference.

How dependent is Cerebras on OpenAI?

Very. Recent coverage says OpenAI accounts for a significant share of Cerebras revenue, which is why any news of OpenAI using Nvidia hardware moves the stock so sharply. Revenue growth from other customers is the most important metric to watch.

What are the fundamentals behind CBRS right now?

Q2 revenue was about $180 million, up 74% year over year, with $25.4 billion in remaining performance obligations as of June 30. About 238 million shares are outstanding. The business is growing quickly. The debate is over what multiple that growth deserves, given the concentration risk and pressure from Nvidia.

Should we buy Cerebras stock at the post-IPO low?

That depends on time horizon and risk tolerance, and this is not financial advice. Short-term traders are watching whether the stock holds through the remaining lockup releases. Long-term investors should focus on Q3 customer diversification and whether Cerebras keeps a measurable speed lead over Nvidia’s Rubin-generation systems.